# Stop re-explaining your business to the AI A model knows what a bounce rate is. It does not know that your March spike is the annual report, that Indian organic traffic is mostly bots, or that you report on custom channels. Give it that once, per property, and every conversation starts informed. [Start free](https://app.aidata.stream/register) [Set up property context](/docs/property-context/) ## What the AI knows before you type Four kinds of context, each used differently. Two are in front of the AI from the first word; one is searched when a question needs it; one is checked when an analysis begins. All of it is shared by the team and scoped to the property. - Presets say how to think. Property context says what this site is. Memory holds what your team has said. The knowledge base holds the detail. - Reports get the same property context as chat, manual or scheduled. - Your own AI client gets it too, through the connector. New conversation on this property before your first message 1. Prompt preset How to think: analyst, SEO specialist, executive summary always 2. Property context What the site is, the conversion that matters, what to ignore always 3. AI memory Facts your team stated in earlier chats, newest first always 4. Knowledge base Entry titles up front; full entries fetched when a question calls for them on demand 5. To-do list Open tasks, checked at the start of an analysis checked Ready. Ask your question; nothing above has to be repeated. [ Always on ## Property context A short briefing the AI reads at the start of every conversation and every report on the property: what the site is, the conversion that matters, traffic to ignore, competitors, seasonality. Set once in Property Settings; the next message picks it up. Every plan. How it works](/docs/property-context/)[ Saved as you talk ## AI memory State a durable fact in conversation and the AI keeps it: your fiscal year, a naming rule, the KPI that counts. One pool per property, shared by the team, newest fact wins. It never saves numbers from your data or anything personal. Incognito mode reads and writes nothing. 20 per property on Free, unlimited on Pro and Business. How it works](/docs/ai-memory/)[ Searched on demand ## Knowledge base Longer entries in Markdown for anything that needs a paragraph: data quirks, custom channel definitions, seasonal patterns, competitor context. The AI sees the titles up front and fetches the full entry when a question calls for it. Drafts stay hidden until published. Every plan. Semantic search when the team has an OpenAI provider, keyword search otherwise. How it works](/docs/knowledge-base/)[ Actions, not facts ## To-do list A shared list per property for the follow-ups analysis turns up. The AI checks it when an analysis starts, adds items on its own when a conversation finds something, and ticks them off when you say so or the data shows it. Due dates, stars, notes, drag to reorder. 25 open tasks per property on Free, 500 on Pro and Business. How it works](/docs/todos/) ## Written down while you talk Say a fact and the AI keeps it. Say "update that" and it edits the existing memory instead of adding a duplicate. Say "forget" and it is gone from the next message. Follow-ups work the same way: ask for a to-do, or let the AI add one when a conversation turns something up. - Every memory carries the date it was last stated, so newer facts win when things change. - Tasks the AI added are marked, and link back to the chat they came from. - The AI completes a task only when you say it is done or the data clearly shows it, and never deletes one unless you ask. - No analytics figures, no fetched content and nothing personal ever goes into memory. Analytics Chat GA4 GSC Memory updated Added to to-do list Remember that our fiscal year starts in April. Saved. From now on I’ll treat April to March as your fiscal year on this property, for everyone on the team. Add a to-do to re-check organic traffic on 1 November. Added “Re-check organic traffic” to the list, due 1 November, marked as added by AI. Nothing is stored without you noticing: the chip appears in the header the moment it happens. ## How the AI finds the right entry The AI sees your entry titles up front and fetches a full entry when a question calls for it. How it decides which entry depends on one thing: whether your team has an OpenAI provider. - With one, search works by meaning. Each published entry gets an embedding, a numerical fingerprint of what it says, generated in the background and refreshed when the title, content or summary changes. A question about numbers jumping every spring can find the entry about the March report spike with no words in common. - Without one, search works by keyword. The words in your question are matched against each entry’s title, content, summary and tags, so the wording has to overlap. It works, but a well-chosen title and tags matter more. - Add an OpenAI provider later and a Generate All Embeddings button backfills every published entry. Entries without an embedding keep working by keyword; each one shows which mode it is in. - AI memory uses the same search when a property holds more memories than fit in a conversation. Knowledge base search Why do our numbers jump every spring? By meaning OpenAI provider on the team - March Traffic Spike – Annual Industry Report traffic, seasonal, reports - India Organic Traffic is Mostly Bots bots, traffic, data-quality - Custom Channel Groupings channels, attribution, ga4 Found with no words in common: the embedding matched the meaning. By keyword no OpenAI provider - March Traffic Spike – Annual Industry Report traffic, seasonal, reports - India Organic Traffic is Mostly Bots bots, traffic, data-quality - Custom Channel Groupings channels, attribution, ga4 Nothing matches: no entry contains “spring”, “jump” or “numbers”. The spike stays hidden until the question uses its words. ### Everything is editable The Memories page lists every saved fact with its date and who saved it: add, edit, delete one, or clear all. Knowledge base entries have drafts, tags and an AI-written summary. Property context is a Markdown field in Property Settings. ### Presets set the role Team owners and admins write as many prompt presets as they like, from a blank system prompt or one of eight templates. Pick one above the message box and the AI approaches the data from that perspective. [More about the chat](/features/ai-chat/). ### On every plan - Property context, knowledge base and presets: every plan - AI memory: 20 per property on Free, unlimited paid - Open to-dos: 25 per property on Free, 500 paid [Compare plans](/pricing/) ## Questions ### What will the AI never put in memory? Numbers from your analytics data, content fetched from your connected tools, and sensitive personal information. It saves only facts you state yourself, and shows a Memory updated chip whenever it does. ### Can I ask a question without the saved context? Yes. Toggle Incognito in a conversation’s header and that conversation neither reads nor saves memories. A property can also switch memory off entirely in its settings; existing memories are kept but no longer used. ### Who can see memories and the to-do list? Everyone with access to the property. Memories are one pool per property so answers are consistent whoever asks. Editing the to-do list needs write permission, for you and for the AI acting on your behalf. ### Property context or knowledge base? Property context is included in every chat and report, so keep it to the always-relevant essentials. The knowledge base is searched when a question calls for it, so it can grow large. Most properties use both. ### Do I need an OpenAI key for the knowledge base? No, but it changes how entries are found. With an OpenAI provider, entries carry embeddings and the AI searches by meaning; without one it matches the words in your question against the title, content, summary and tags. Both modes can coexist on one property, and adding a key later backfills the embeddings with one click. ### What happens when a Free property hits its memory limit? The AI can still update existing memories; it just cannot add new ones until you delete some or upgrade. The to-do list behaves the same way at its limit, and the AI tells you what it wanted to add. ### Does this work from my own AI client? Yes. Through the connector, the AI can read the knowledge base, save memories where memory is enabled, and add or complete to-dos, with the same permission rules as the built-in chat. ## Launch history - 2026-10-16 Shipped [Never lose a follow-up again: the to-do list](/blog/your-ai-analyst-now-keeps-a-to-do-list/) - 2026-07-19 Shipped [Your AI analyst now remembers: AI memory](/blog/your-ai-analyst-now-remembers/) - 2026-05-27 Shipped [Property context: always-on background for the AI](/blog/property-context/) - 2026-04-16 Shipped [Property knowledge base](/changelog/) Page last updated 2026-10-25. ## Brief the AI once Write a paragraph of property context, then let memory and the to-do list fill in the rest as you work. [Start free](https://app.aidata.stream/register) [Knowledge base examples](/docs/knowledge-base/) --- Source: https://aidata.stream/features/business-context/